k = np.load(yfile) return (j, k) # Load data and normalize X, y = load_data() # Change type and Normalize X = X.astype('float32') X /= 255 # 1-hot encoding y = np_utils.to_categorical(y, num_classes) skf = StratifiedKFold(n_splits=4,random_state=34, shuffle=True) print('using K-cross validation with %s folds' % skf.get_n_splits(X, y)) # Model model = Sequential() def define_model(): print('\n defining model . . .') gaussian = RandomNormal(mean=0., stddev=0.1) cons = Constant(value=2.) model.add(Conv2D(filters=32, kernel_size=(7, 7), padding='same', kernel_initializer=gaussian, use_bias=True, bias_initializer=cons, bias_constraint=maxnorm(5.),input_shape=X_train.shape[1:])) model.add(Activation('relu')) model.add(BatchNormalization(axis=-1, momentum=0.99, epsilon=0.001, center=True, scale=True)) model.add(MaxPooling2D(pool_size=(3,3)))